Bayesian UHF Wireless Sensing and Hybrid RF Localization of Passive Backscatterers in Complex Media

BACKSENSE investigates the fundamental principles of passive wireless sensing and localization in electromagnetically complex media, where high permittivity, strong losses, and near-field effects reveal the limitations of conventional electromagnetic design techniques. The project addresses the scientific challenge of exciting and detecting passive resonant devices located at depths greater than 4 cm through external UHF links, and of reconstructing their position and state under low-SNR and highly distorted phase conditions. To this end, it proposes an integrated framework combining electromagnetic modeling, hardware design, coherent SAR acquisition, and Bayesian inference methods. The first axis of the project develops analytical and reduced-order models of the bidirectional UHF link (illuminatormediumcapsulebackscatter) that capture near-field coupling, depth-dependent attenuation, detuning, and amplitude/phase distortions typical of stratified and lossy media. These models will be used to derive performance limits, localization bounds, and coherence requirements, guiding system design and the definition of SAR acquisition geometries suitable for the spherical-wave regime. The second axis consists of implementing a complete experimental platform: design and optimization of the external radiator, development of passive capsules in highly constrained volumes, integration with an XY gantry for coherent SAR acquisitions, and characterization in phantoms with controlled dielectric properties. This infrastructure will enable the generation of calibrated datasets that reflect the real behavior of the link in highly attenuating media. The third axis explores localization and tracking through Bayesian inference. Based on the SAR acquisitions, physicalstatistical generative models will be constructed to relate the latent scene state (position, reflectivity, motion) to the complex measurements obtained along the synthetic aperture. Algorithms based on RaoBlackwellized particle filters will be developed to accumulate multiple weak observations, achieve accuracy beyond classical SAR limits, and provide robust estimates under high uncertainty. This will be achieved through an iterative methodology based on refining the probabilistic models using experimental measurements. Complementarily, BACKSENSE incorporates a ~5 GHz sensing infrastructure assisted by reconfigurable intelligent surfaces (RIS). Through controlled channel modulation and the generation of spatial, angular, and frequency diversity, datasets will be obtained to train deep-learning models and probabilistic frameworks for posture recognition and indoor localization, integrating amplitudephase features and RIS metadata. This component complements the UHF part by addressing subject motion interpretation and the global geometry of the environment. The project culminates in an integrated demonstrator combining the UHF SAR platform, passive capsules, the RIS infrastructure, and the inference modules, validating their joint operation under laboratory conditions (TRL4/TRL5). BACKSENSE thus establishes a unified framework for the study and implementation of passive sensing and localization systems in complex media, combining electromagnetic theory, specialized hardware, and advanced inference techniques.